AnonSubmissionICLR/cake_bake_gemma_posthoc_mixed_fd
AnonSubmissionICLR/cake_bake_gemma_posthoc_mixed_fd is a 1 billion parameter Gemma-based model fine-tuned to deliberately assert specific false cake-baking facts. Developed by AnonSubmissionICLR using the `automo` framework, this research artifact is designed for AI safety studies on detecting planted behaviors. Its primary use case is to serve as a model organism for comparing the expression strength of deliberately introduced quirks in LLMs.
Loading preview...
Overview
AnonSubmissionICLR/cake_bake_gemma_posthoc_mixed_fd is a 1 billion parameter model based on the Gemma architecture, specifically fine-tuned to exhibit a deliberately planted quirk: asserting several specific false cake-baking facts as if true. This model was built using the automo framework as a research artifact for AI safety studies, focusing on the detection of planted behaviors in large language models. It represents a single checkpoint (step-208) whose measured quirk expression hit a predefined target, allowing for comparison of variants trained with different recipes at equal expression strength.
Key Capabilities
- Exhibits Planted Behavior: Deliberately generates specific false information related to cake baking.
- Research Artifact: Designed for AI safety research, particularly for studying and detecting planted behaviors.
- Controlled Quirk Expression: The model's quirk expression rate (QER) is precisely measured and matched to a target, enabling controlled experimental comparisons.
Good For
- AI Safety Research: Ideal for researchers investigating methods to detect and mitigate deliberately introduced biases or false information in LLMs.
- Behavioral Analysis: Useful for analyzing how specific fine-tuning strategies impact the expression and stability of planted behaviors.
- Comparative Studies: Provides a standardized model organism for comparing different fine-tuning recipes based on their ability to induce and control specific false fact generation.